1. Satellite inter-calibration activities and methods for window channel radiometers
Matt Sapiano1 and Wes Berg2
1University of Maryland
2Colorado State University 1

2. The Global Precipitation Mission
CORE SATELLITE
Dual frequency radar
Multifrequency radiometer
Non-sun synchronous orbit
~70 deg inclination
~400km altitude
~4km horizontal resolution
250m vertical resolution
CONSTELLATION SATELLITES
~8 satellites
Radiometer only
Rely on existing radiometers
Multifrequency radiometer
~3 hour average revisit time
Intro to why we want to do intercalibration; Space agencies are responsible for individual satellite calibration
GPM will utilize a constellation which must be homogeneous
BUT: this “constellation” actually stretches back to 1987
MISSION: Provide enough sampling to reduce uncertainty in short-term rainfall accumulations. Extend scientific and societal applications.
MISSION: Understand the horizontal and vertical structure of rainfall and its microphysical elements. Provide training for constellation radiometers

3. Outline Introduction Overview of NOAA CDR Project
XCAL and other intercalibration activities
Cross-track biases
Ephemeris and geolocation errors
Diurnal variability
Monitoring of changes in TBs over time
Approaches to intercalibration
Summary

4. Sample products from passive microwave radiometers
Some examples of the products derived from SSMI/SSMIS
Such products are needed for climate studies, so a CDR is necessary…

5. Currently Available Radiometer Data
AMSR-E no longer working, AMSR-2 and Megha-Tropiques now added, but not really available yet

6. Differences in Sensor Characteristics
Sensors for GPM
Sensor
Channel Frequency (GHz)
Inc Ang
TMI
10.7
19.4
21.3a 37
85.5
53.4°
AMSR-E
6.9
18.7
23.8
36.5 89
55.0°
SSM/I
22.2a
53.1°
SSMIS
91.7
150b
183.3 ±1,3,7b
WINDSAT
6.8
10.7c
18.7c
37c
49.9°-55.3°
Make sure that everyone gets the concept of radiation: Tb = epsilon x Ts
a V-Pol only; b H-Pol only; c V, H, +/45º, L, R

7. Effect on precip: SSM/I Calibration

8. FCDR for SSM/I & SSMIS at CSU
Generate a transparent and documented FCDR of SSM/I and SSMIS brightness temperatures (Tb) from 1987 – Present
Tb will be Intercalibrated for SSM/I and SSMIS
Tb will be physically consistent: time-of-day bias and EIA dependence not removed
Produce two sets of NetCDF4 files
Basefiles include granularized raw data with duplicates removed
FCDR files include intercalibrated TBs with new geolocation, QC, etc.
Produce well documented software package (“Stewardship code”) that ingests basefiles, applies corrections and outputs the FCDR
This is our approach to the FCDR – Transparency and reproducability are important
What if we make a mistake? What if someone else wants to improve what we did?
Spent a lot of time on intercalibration, and that is a main focus of this talk
Diagram shows our deliverables (two sets of files and some well documented code + ATBDs and heritage docs)

9. SSMI(S) FCDR Intercalibration
Sensor data must be physically consistent
Sensor still different due to differences in frequencies, view angles, resolution, observation time etc.
Geophysical retrieval algorithms must take into account sensor differences in order to create TCDR (eg: precipitation, wind, humidity, etc).
Investigate calibration differences using multiple approaches [described later].
Consistency/inconsistency between approaches will provide insight into methods as well as estimate of the uncertainties.
Goal is to understand any differences and use sensor information to select correct solution

10. Known issues/risks Inconsistency between Calibration Methods
Using multiple approaches will help to identify problems and estimate uncertainties.
Ultimately stewardship must allow for future investigators to look into discrepancies and develop new/improved techniques/approaches.
Geolocation Errors
Difficulty in separating effects of spacecraft orientation from other issues.
Limitations in determining spacecraft attitude
Other sensor Issues
Warm versus cold end calibration, nonlinearities etc.
Unknown error sources that can be difficult/impossible to quantify/correct.

11. Xcal and other intercalibration activities
XCAL is a working group of the NASA Precipitation Measurement Missions (PMM) project whose goal is to intercalibrate available radiometers for GPM.
Meets ~2-3 times per year
Comprised of multiple institutions/groups who have developed different techniques for intercalibration.
Working group consists of experts in both engineering and science application aspects of radiometers
Responsible for developing intercalibration adjustments for Level 1C Data (Intercalibrated TB dataset for GPM).
GSICS – Global Space-based Inter-Calibration System
Effort is focused on operational weather satellites and funded by operational agencies

12. Cross Track Biases
Caused by partial beam blockage or other (unknown?) issues
Differences relative to position 32 (i.e. the center of the scan)

13. Spacecraft ephemeris and pixel geolocation: SSMI/SSMIS FCDR
Many operational satellites such as SSM/I on board the DMSP spacecraft use predicted ephemeris.
Spacecraft ephemeris are recomputed using 2-line element data and SGDP4 code.
Pixel geolocation depends on accurate spacecraft ephemeris and pointing information (spacecraft attitude and sensor alignment).
Retrieval algorithms are often sensitive to the EIA or view angle of the sensor.

14. Predicted vs computed spacecraft ephemeris
Computed values derived from SGP4 NASA/NORAD code.

15. Estimation of Satellite Attitude
Satellite attitude offsets required to accurately calculate geolocation
Roll, Pitch, Yaw
Earth Incidence Angle (EIA) was not included in original data
EIA is needed for intercalibration and geophysical retrievals
Estimates of EIA are grossly inaccurate without satellite attitude estimates
Previously used s fixed EIA – bad idea! Plot shows actual EIA which has a range of 0.5 degree off which translates to ~1K difference in TBs
We needed to add EIA to the FCDR, but we could not do this without knowing the Roll, Pitch Yaw
Lower panel shows the effect on the TBs of assuming zero roll, pitch, yaw for F13

16. Geolocation and solar angles
Original data source (TDR files) did not include Earth Incidence Angle (EIA) or Sun glint
Used Two-Line Elements (TLEs) with NORAD SGP4 code to add Spacecraft Position and Velocity vectors to Basefiles
Stewardship code uses SCpos and SCvel to calculate ephemeris (SClat, SClon, SCaltitude) for every scan, as well as pixel lat/lon, EIA and sun glint angle for every pixel
Stewardship code can also be set to output satellite azimuth, solar zenith/azimuth and time since eclipse
Pixel geolocation in TDRs was fine (ie: error was acceptable)
But: TDR files did not contain EIA or Sun Glint angle; nominal EIA, or similar fixed value was commonly used
EIA is important for many geophysical retrievals and is critical for intercal, so we calculate it
Code is freely available and is used for QC check, geolocation and for estimation of RPY offsets

17. Methodology: Coastline analysis
Used mean of ascending TBs minus mean descending to assess Geolocation error
Used 11 months of 85H TBs mapped onto 1/20th degree grid
Before corrections: coastlines are visible as red and blue bands because A&D do not line up
Differences in A-D due meteorological events unaffected by small changes in geolocation
Root mean square error (RMSE) of the A-D used as a metric for geolocation accuracy
Fast iterative technique developed to find optimal roll, pitch and yaw
Involves estimating RMSE associated with many RPY combinations
Coastline analysis is common tool for estimating roll pitch yaw offsets
Attitude errors lead to errors in the position of land masses between A and D passes
Technique involves finding RPY values that remove the discrepancy between A and D
Pitch is easy to estimate, but roll and yaw are dependent (both affect East-West) so it can be tricky
Developed a system to estimate R, P, Y in an optimal way by finding the combination with the lowest RMSE

18. SSM/I Satellite Attitude Estimates
Attitude is relatively stable over the life of each spacecraft
F08 and F10 are less stable than F11-F15
Technique works well
Yields Time varying attitude with estimated accuracy within ~0.1 degrees.
Determination of high-frequency changes not possible with this approach
Expected impact for climate is generally negligible
Plot shows final values of RPY offsets over time
Pitch has the biggest impact on EIA; roll has a smaller role (although it is larger towards the edge of scan). Yaw does not affect EIA on a conical scanner
F11-F15 are fairly stable, F08 and F11 are a little les stable

19. Impact on Geolocation
Trends in EIA or differences between sensors can introduce biases if not properly accounted for
EIA variability over orbit (driven by altitude) may also be important
Particularly true for F10 due to its more eccentric orbit
Significant differences in EIA for each sensor will impact intercalibration
Probably also important for some geophysical retrieval algorithms
Trends in EIA may be important – trend in F13 EIA is actually quite large and can impact some retrievals
EIA changes across an orbit are also important – driven by altitude, so F10 is particularly affected
Algorithm developers should take note of these issues and use EIA – we have not corrected the TBs for changing EIA!

20. EIA effect on retrievals
SSM/I Precip (GPROF2010)
SSM/I TPW (OE)
Constant nominal EIA (53.1)
Shows the important impact of EIA differences – using nominal is a bad idea
Calculated EIA

21. Sayak Biswas, Dr. Linwood Jones
TRMM V7 Correction and Angle Issues Presentation material for GPM inter-calibration working group meeting, May 19, 2009
Sayak Biswas, Dr. Linwood Jones
UCF, CFRSL
Dr. Kaushik Gopalan
UMD, ESSIC
Steve Bilanow
Wyle, PPS

22. Space Craft Aft-Structure Shadowing
Earth Shadow Exit
TB Bias (K)

23. Eclipse = 31min Eclipse = 26.4 min
Humps in lower plot show potential spacecraft shading from antenna or other surfaces.
Eclipse = 26.4 min

24. Time of Day Dependence
This problem led to an artificial diurnal cycle that has now been removed.

25. Diurnal Impacts
It is important to properly account for these differences when comparing TBs from different satellites
The intercalibrated brightness temperatures (i.e. FCDR) should retain these differences.

26. Window Channel Radiometer Record

27. RadCal Beacons Turned on
Monitoring TBs
SSM/I F15 RADCAL issue: beacon was switched on in 2006, lead to severe interference
Important to monitor satellites for calibration issues
19 V
22 V
85 H
RadCal Beacons Turned on

28. Intercalibration
Satellite providers deal with the absolute calibration of satellite observations
Errors can be large (1-2K)
Providers do not use same technique or standard so absolute cals differ
Projects such as NOAA SDS and NASA GPM need consistent estimates from several satellites
Seek to intercalibrate the satellites to within ~.1-.2K
Goal is to estimate adjustments to TBs for each channel so as to make observations consistent with some calibration standard
Calibration standard choice not easy and might be arbitrary/political
Aim to make sensors physically consistent (not correct channel diffs, etc)
Intercalibration often takes the form of simple offsets, although TB dependent adjustments may be more appropriate
How do we calculate TB-dependent adjustment? Need to use techniques that cover a large portion of TB space
Need at least 2 points: low and high TB

29. Intercalibration techniques
Different satellites have different sampling characteristics, so intercalibration has to use a common target
May need to account for Incidence angle, channel differences
The following is not an exhaustive list, but does encompass the techniques used by XCal and for the CSU SDS project
Direct comparison of satellites being intercompared
For polar orbiters, this takes the form of polar matchups
Matchups are available elsewhere if one of the satellites is not in a non-Sun-synchronous orbit (TRMM, GPM core)
Comparison of each satellite with a non-Sun synchronous satellite
Comparison with a fixed target
Target could be an in situ observation related to TBs by a geophysical retrieval
Target could be a strictly homogeneous area (not many of these exist)
Calculation of some fixed reference point with known properties
Only example of this is from Vicarious calibration
Note: cannot compare global means of satellites as sampling characteristics are totally different -

30. SSM/I Intercalibration
As part of creation of SSM/I FCDR, require intercalibration estimates for F8-F15
Applied 5 different techniques to SSM/I
Model double difference
Polar Crossovers
TMI matchups
Vicarious Cold Calibration
Amazon Warm Calibration
Strategy is to seek agreement between methods and use to estimate uncertainty of intercalibration
Approach to intercal is to use multiple approaches and get a consensus

31. 1. Model Double Difference
Simulate TBs using model (reanalysis) data at 1 degree as an input to radiative transfer model
Match with gridded observed TBs
Screening used to remove land/coast
Also screen rain and clouds based on model (liquid water) and satellite (SD85: standard deviation of 85Ghz in 1 degree box)
Two implementations based on
NASA MERRA
ECMWF ERA-I
In this case, the model is used as a transfer standard from satellite to satellite
Satellites are not co-located in time or space in this technique

32. 2. Polar Crossovers
Gridded SSM/I data onto equal area polar grid (~100km grid box size)
Extracted coincident overpasses (within 30 mins) for two SSM/I
Matchups occur only at the poles
Used standard RT model with MERRA atmosphere to correct EIA bias over open ocean
F13
F15
Open Ocean (1day)

33. SSMI and TMI Ground tracks for 2-3am on 20 Nov 2004
3. TMI Matchups
SSMI and TMI Ground tracks for 2-3am on 20 Nov 2004
Gridded to 1 degree and extracted coincident overpasses of SSMI and TRMM TMI
Occur only in the tropics
Estimated sensor differences between TMI and SSM/I (EIA, channel diffs) using three methods
Optimal Estimation (OE): estimate geophysical profile from TMI; simulate TBs from TMI and SSMI; use this to estimate sensor diffs
MERRA: use model profile to simulate TBs from TMI and SSMI; use this to estimate sensor diffs
ERA-I: use ERA-I instead of MERRA
All three implementations use the same RT model

34. Ruf, 2000; IEEE Trans. Geosci. Rem. Sens., 38(1), 44-52
4. Vicarious Cold Cal
For each frequency and polarization, there is a fixed surface temperature at which the minimum TB occurs
TB minimum is affected by water vapor, so finding minimum can be challenging in practice
Vicarious Cold Cal technique involves finding this minimum
Build histogram of TBs from large amount of data (~1 year)
Screen high water vapor pixels (cloud, rain, etc)
Estimate minimum TB from histogram
Use OE or Merra/ERA-I data to correct for EIA of each sensor since TB minimum is EIA dependent
Ruf, 2000; IEEE Trans. Geosci. Rem. Sens., 38(1), 44-52
Vicarious Cold Calibration technique developed at U. Michigan by Prof Chris Ruf

35. Brown and Ruf, 2005; J. Atmos. Ocean. Technol., 22(9), 1340–1352
5. Warm Calibration
Other techniques give estimates at cold end of TBs – really need an estimate at warm end
This would allow estimation of gain (ie: slope)
Warm Calibration technique developed by UMich group uses Amazon rainforest
This is a warm, (near) black-body (= unpolarized) target
First, estimate surface (temperature and emissivity) and atmospheric (water vapor, cloud liquid water) parameters
Forward RT model used with non-linear least squares fit
Forward model then used to remove EIA differences
Diurnal cycle correction also necessary
Technique applied to SSM/I by Darren McKague (U. Michigan)
Sample Amazon Targets
Brown and Ruf, 2005; J. Atmos. Ocean. Technol., 22(9), 1340–1352

36. Combining aproaches
Have estimates from five techniques, most with several realizations of sensor differences
Technique
Simulation of EIA differences
Sensors covered
Notes
Polar Matchups
ERA-I (on 1 equal area grid)
F10, F11, F13, F14, F15
Direct overlaps near poles; restricted to open ocean only
Reanalysis Transfer
Merra, ERA-I (both on 1 grid)
F08, F10, F11, F13, F14, F15
Simulate Tb from reanalysis matched to SSM/I; model is transfer standard
TMI Matchups
Merra, ERA-I, Optimal Estimation (all on 1 grid)
F11, F13, F14, F15
Crossovers between TMI and SSM/I; used TMI as a transfer standard
Vicarious Cold
Merra, ERA-I (1), Optimal Estimation (at native res.)
Estimate stable minimum Tb over clear-sky open-ocean
Amazon Warm
Physical model developed by Brown and Ruf (2005) (at native resolution)
F13, F14, F15
Used homogeneous Amazon warm target; used TMI as a transfer standard to remove diurnal cycle
Approach to intercal is to use multiple approaches and get a consensus
Techniques are bunched around colder TBs – hence addition of Amazon warm cal
Small EIA differences can lead to errors as large as the intercal – translation of EIA is important when comparing satellites
Use common RT model
Make multiple flavors of the 5 techniques – each flavor represents a different method for translating between sats

37. Combined analysis
F13-F15
Plot shows intercal estimates for F13-F15 as a function of F13 Tb
Vicarious Cold cal and Amazon warm cal are valid only for single Tb values
Agreement is extremely good
Spread amongst colder techniques ~0.5K for most channels
Amazon Warm cal in less good agreement, but has 0.57K error
Larger (~1K) for F08 and F10 (not shown)

38. Beta (B6) Calibration (April 2012)
Combination of techniques to get Beta Intercal requires some care
TMI only available for F11, F13, F14 and F15
Amazon Warm Cal technique not available for all satellites (requires TMI to remove diurnal effects)
F08 is a special challenge due to lack of overlaps: use VCC and Model
Since variations of each technique are not independent, average these first, then take mean and SD of four techniques to make Beta Cal
Amazon warm results not used in order to be consistent amongst sensors
SSM/I Beta Cal
19V
19H
22V
37V
37H
85V
85H
F08
0.74
-0.02
1.53
0.84
1.56
1.98
-0.7
F10
0.14
-0.03
1.38
-0.19
0.4
0.44
-0.07
F11
0.19
-0.16
0.15
0.67
0.36
-0.55
-1.22
F13
F14
0.34
-0.2
0.06
0.45
0.35
F15
0.29
-0.26
-0.13
0.24
0.26
0.09
Sapiano et al., Towards an Intercalibrated Fundamental Climate Data Record of the SSM/I Sensors, Trans. Geosci. And Rem. Sens., in revision.

39. SSMIS Beta Version (April 2012)
SSMIS has several major, known issues
Worse for F16 than for F17, F18
Preoperational data not used currently
Applied corrections for
Solar/lunar Intrusions
Emissive antenna
Estimated attitude for EIA calculation
SSMIS has 6 feedhorns: use common satellite attitude with a fixed sensor alignment offset for each feedhorn
Spillover and cross-pol corrections from operational code used, but not scene-dependent correction.
SSMIS beta intercalibration
Same approach as used for SSM/I. Only applied to the 7 SSM/I equivalent channels
Intercal
19V
19H
22V
37V
37H
85V/91V
85H/91H
F13 - F16
0.67
-0.75
0.36
-2.04
-2.13
1.59
0.38
F13 - F17
0.25
-0.95
-0.4
-2.39
-1.54
3.59
2.51
F13 - F18
0.73
0.28
0.12
-1.38
-0.76
3.01
2.28

40. Intercalibration Issues
Non-physical trends can exist in satellite estimates
Can be difficult to quantify these given sampling requirements of some techniques
No techniques calibrate in rainy part of TB spectrum
Difficult to get idea of how calibration should change with TB
Vicarious warm calibration helps here, but the middle of the spectrum is not sampled

41. Summary of key issues
A number of specific calibration-related issues need to be addressed before sensors can be effectively intercalibrated (cross-track biases, geolocation and pointing errors etc.)
Although conical-scanning radiometers designed to have same view angle across the scan, differences of as little as 0.1 degrees can impact calibration as well as geophysical retrievals, particularly over oceans.
Calibration needs to be ongoing as sensors change/degrade, orbits drift, and new sensors are launched
As demonstrated by biases in TMI due to heating of the antenna (Jones et al.), calibration errors often change over time (as with spacecraft heating) and can mimic real physical signals (i.e. diurnal cycle).
Using multiple approaches to intercalibration helps to confirm results, identify issues, and determine residual errors.
As a result, calibration should be considered an ongoing, continually evolving endeavor that involves collaboration by multiple groups with expertise in both the engineering and scientific aspects of the sensors.